Review article

Forecasting of the construction completion based on the modeling of nonlinear dependence on delays of individual works

UDC: 

69.003:658.012.22

DOI: 

10.23968/1999-5571-2022-19-2-83-90

Pages: 

83-90

Annotation: 

Methods of forecasting construction duration can be divided into two classes: firstly, the extrapolation methods, the essence of which is to transfer the results of the current state of the system to its future state based on taking certain characteristics as invariants, and secondly, forecasting methods based on regression determination of the functional relationship between delays in performing individual works and the total construction delay. These methods can be implemented using two different models: multiple regression analysis and neural network modeling. However, to choose one of the models, it should be kept in mind that the accuracy of the prediction is affected by the non-linearity between the total construction delay and the delays of performing individual works. The presented study has shown that nonlinearity is a constant attribute of the estimated work schedules, as demonstrated by the example of calculating a schedule based on the principles of stream-type construction organization. A comparison of the linear regression analysis apparatus and the neural network modeling apparatus showed that the average values of the relative deviations for the two-layer and three-layer perceptrons become smaller than the values calculated using multiple linear regression. And for the three-layer perceptron, these values are guaranteed to be lower than similar values obtained using the multiple linear approximation apparatus.

Authors: 

Bolotin S. A. Saint Petersburg State University of Architecture and Civil Engineering St. Petersburg, Russia

Al-Janabi M. A. Saint Petersburg State University of Architecture and Civil Engineering St. Petersburg, Russia

Bohan Kh. A. Saint Petersburg State University of Architecture and Civil Engineering St. Petersburg, Russia

Links to authors' articles: 

Journal Issue